Research interests

Distance-based and kernel-based methods

• Clustering and instance based learning
  • AI Comm (2003)
  • Relational instance based regression (ICML'03)
  • Instance based function learning (ILP'99)
  • Logical clustering trees (ICML'98)
• Defining metrics
  • Metrics on sets: Acta (2001),
  • Metrics on first order logic: ILP'98, NAIC'98
  • Distance measures between atoms (1998)
• Kernels
  • Graph kernels MLJ (2004), ILP'03
  • Expressivity versus efficiency of graph kernels (MGTS'03)

Frequent pattern mining

• Condensed representations for Inductive Logic Programming: KR'04, ILP'04
• Efficient theta-subsumption of sets of patterns
• Frequent pattern mining under generalized subsumption (DBDBD'04)
• On efficient mining of compactly represented sets of frequent patterns in relational languages (IKB'04)

Efficient data mining and complexity issues

• Compact representation of knowledgebases: MLJ (2004), ILP'02, FLF'03 BNAIC'02
• Query packs: JAIR'02, ILP'00
• Polynomial time metric on sets based on matchings (Acta 2001)
• Efficient theta-subsumption of sets of patterns
• Expressivity versus efficiency of graph kernels (MGTS'03)

Active learning, reinforcement learning and planning

• Searching for compound goals using relevancy zones
• Active learning: The domain expert is not an oracle
• Incremental regression algorithms for relational reinforcement learning
  • Gaussian processes: MLJ, ILP, ICML-RRL
  • Relational Instance based regression: ICML, BNAIC'03
  • Decision trees
• Multi agent
  • Relational multi agent learning: Lamas,
• Efficiency
  Convergence when using tree learning
  • ICML-RRfRL, FLF
• The game of go
  • Tsume-go move-ordering learning: CG, BNAIC
  • Searching for compound goals using relevancy zones
  • Opponent modeling by analysing play
  • Other: Challenges, Learning techniques
• Informed reinforcement learning: ICML-RRL, FLF

Bayesian methods for relational data

• Probability trees: ECML, ILP
• Logical Bayesian networks: ILP MRDM, BNAIC
• Domain and combining rules in Bayesian logic programs
• Using the structure of generality lattices for belief networks: ILP, MSL, BNAIC
• Multi instance neural networks

Applications

• HIV data mining (collaboration with Rega Institute and South-Africa)
• Intensive care data mining (collaboration with University Hospitals Leuven).
• Active learning in coherent laser control (collaboration with Computational biology group Aberystwyth (UK))
• First order models for the predictive toxicology challenge

Various

• Generalizing refinement operators to learn prenex conjunctive normal forms
• Hierarchical multi-classification